Experience

Unmesh Mashruwala Innovation Cell

• India
• Research Services
• 1 - 100 Employee

• Team Lead

  • Mar 2022 - Present

  I'm leading the technical forefront of a multidisciplinary team of 40 students developing aerial autonomy solutions for aerial robotics competitions around the globe. I steer the research within different subsystems (viz. control, motion planning, and vision) and plan the overall undertakings of the team for the year. I'm also leading our recent collaboration with WWF-India for wildlife preservation, for which the team is providing aerial surveillance solutions.

• Control Subsystem Head

  • Mar 2021 - Mar 2022

  Head of the Controls subsystem of the team AeRoVe, leading the ideation and development of control systems and strategies for all the team's endeavors. Led the development of a robust trajectory-tracking framework for Quadrotors. Developed a cascade position and attitude controller based on PID and LQR control and implemented an Integral Sliding Mode Controller for mitigating external disturbances.

• Junior Engineer

  • Aug 2019 - Apr 2021

  As a Control Engineer, I was responsible for developing a central autopilot system (FSM) for decision-making & autonomy. I simulated the IARC mission 9 in Gazebo using ROS with PX4 and ArduPilot as the flight control stack. I used ROS Joint-State-Controller to autonomously carry out in-flight manipulation of the custom robotic gripper. As a Perception Engineer, I implemented Visual-Inertial SLAM algorithms for real-time pose estimation of a quadrotor. I deployed ORB-SLAM2 on Intel D435… Show more As a Control Engineer, I was responsible for developing a central autopilot system (FSM) for decision-making & autonomy. I simulated the IARC mission 9 in Gazebo using ROS with PX4 and ArduPilot as the flight control stack. I used ROS Joint-State-Controller to autonomously carry out in-flight manipulation of the custom robotic gripper. As a Perception Engineer, I implemented Visual-Inertial SLAM algorithms for real-time pose estimation of a quadrotor. I deployed ORB-SLAM2 on Intel D435 RGBD camera for pose estimation with 5cm in-flight accuracy. I also optimized YOLOv4 using TensorRT engines to speed up inference by 300% on Jetson Xavier.



Mitacs

• Canada
• Research Services
• 400 - 500 Employee

• Mitacs Globalink Research Internship

  • May 2022 - Dec 2022

  Worked on Multi-Agent Reinforcement Learning under Prof Malek Mouhoub at the Computer Science Department, University of Regina. I was responsible for developing a multi-robot solution for a field exploration problem. Initially, I experimented with classical algorithms and observed a suboptimal exploration since they don’t account for the intents of other agents. To address these problems, I proposed a MARL-based approach for coordinated goal selection. I designed an LSTM-based actor-critic… Show more Worked on Multi-Agent Reinforcement Learning under Prof Malek Mouhoub at the Computer Science Department, University of Regina. I was responsible for developing a multi-robot solution for a field exploration problem. Initially, I experimented with classical algorithms and observed a suboptimal exploration since they don’t account for the intents of other agents. To address these problems, I proposed a MARL-based approach for coordinated goal selection. I designed an LSTM-based actor-critic architecture to effectively handle partial observability and to achieve team size flexibility while training decentralized policies. Moreover, to encourage exploration, I generated intrinsic team rewards inspired by Random Network Distillation based on the uniqueness of an encountered state. The proposed approach reduced the exploration time by over 50%, and the close interactions between agents were drastically reduced compared to the classical Nearest Frontier approach. The work is under review at the International Conference on Unmanned Aerial Systems (ICUAS) 2023. Show less Worked on Multi-Agent Reinforcement Learning under Prof Malek Mouhoub at the Computer Science Department, University of Regina. I was responsible for developing a multi-robot solution for a field exploration problem. Initially, I experimented with classical algorithms and observed a suboptimal exploration since they don’t account for the intents of other agents. To address these problems, I proposed a MARL-based approach for coordinated goal selection. I designed an LSTM-based actor-critic… Show more Worked on Multi-Agent Reinforcement Learning under Prof Malek Mouhoub at the Computer Science Department, University of Regina. I was responsible for developing a multi-robot solution for a field exploration problem. Initially, I experimented with classical algorithms and observed a suboptimal exploration since they don’t account for the intents of other agents. To address these problems, I proposed a MARL-based approach for coordinated goal selection. I designed an LSTM-based actor-critic architecture to effectively handle partial observability and to achieve team size flexibility while training decentralized policies. Moreover, to encourage exploration, I generated intrinsic team rewards inspired by Random Network Distillation based on the uniqueness of an encountered state. The proposed approach reduced the exploration time by over 50%, and the close interactions between agents were drastically reduced compared to the classical Nearest Frontier approach. The work is under review at the International Conference on Unmanned Aerial Systems (ICUAS) 2023. Show less



Aeromodelling Club IIT Bombay

• India
• Airlines and Aviation
• 1 - 100 Employee

• Institute Technical Convener

  • Mar 2020 - Mar 2021

  Spreading drone culture among the students by organising various competitions and mentoring them. I have conducted the ITSP Bootcamp on the design, control, and automation of aerial vehicles. I have organized the PlutoX hackathon workshop on drone programming in collaboration with Drona Aviation. I also formulated and conducted the inter-hostel Aeromodelling GC involving path planning and PID control. I was one of the two conveners who organized and conducted the first edition of TriviaJet, the… Show more Spreading drone culture among the students by organising various competitions and mentoring them. I have conducted the ITSP Bootcamp on the design, control, and automation of aerial vehicles. I have organized the PlutoX hackathon workshop on drone programming in collaboration with Drona Aviation. I also formulated and conducted the inter-hostel Aeromodelling GC involving path planning and PID control. I was one of the two conveners who organized and conducted the first edition of TriviaJet, the annual aviation quiz of IITB with 100+ participants Show less Spreading drone culture among the students by organising various competitions and mentoring them. I have conducted the ITSP Bootcamp on the design, control, and automation of aerial vehicles. I have organized the PlutoX hackathon workshop on drone programming in collaboration with Drona Aviation. I also formulated and conducted the inter-hostel Aeromodelling GC involving path planning and PID control. I was one of the two conveners who organized and conducted the first edition of TriviaJet, the… Show more Spreading drone culture among the students by organising various competitions and mentoring them. I have conducted the ITSP Bootcamp on the design, control, and automation of aerial vehicles. I have organized the PlutoX hackathon workshop on drone programming in collaboration with Drona Aviation. I also formulated and conducted the inter-hostel Aeromodelling GC involving path planning and PID control. I was one of the two conveners who organized and conducted the first edition of TriviaJet, the annual aviation quiz of IITB with 100+ participants Show less